Pattern spectra are a powerful tool for characterizing image content at both local and global levels. They have been successfully applied to image or patch retrieval, enabling efficient processing of large satellite imagery, for instance. In Earth Observation, multitemporal imaging, or satellite image time series, provides not only spatial but also temporal information. In this context, retrieval aims to identify similar spatio-temporal patterns. In this paper, we address the problem of efficient content-based time series retrieval using pattern spectra. To this end, we revisit the pattern spectra retrieval methodology originally defined in the spatial domain and adapt it to the spatio-temporal case. Preliminary results on the SECOND dataset demonstrate its effectiveness compared to both a deep learning baseline and a uniform guessing baseline, especially when using small patch sizes.